|
Following
CHI’s Structure-Based Drug Design (SBDD) meeting this
year, leaders in the field joined Pharma DD
editor-in-chief Malorye A. Branca and CHI producer
Shelly Amster to discuss the major trends influencing
SBDD, including where the field stands, how technology
is changing it, and whether sharing
“pre-competitive” data could lead to further
acceleration. Below
is an excerpt from that discussion.
The
panelists included:
Sean Ekins, vice president of Computational
Biology, ACT; Klaus Müller, head of Science and
Technology Relations at Roche; Mark Murcko, vice
president and chief technology officer at Vertex; and
Tomi K. Sawyer, senior vice president, Drug Discovery,
ARIAD Pharmaceuticals.
Mark
Murcko: One thing
that really stands out at these meetings, is the
dichotomy between the people who are trying to make
drugs and those who have a particular technology they
are championing. The
latter group tends to present their technology as the
solution to the most critical part of drug
discovery. It’s
a bit annoying because we know that making a drug
requires so many different components to be
successful. I
don’t care how good one technology is — it is not
the whole story. So
we have to get across the idea that the whole is greater
than the sum of the parts.
People have to think about how all the components
fit together, and how one experiment drives the next.
You need a wide variety of tools.
Klaus
Müller:
I have to echo Mark’s comments, but I think
part of the problem is that a lot of these people just
don’t have access to the really interesting data, and
maybe that’s partly our responsibility to help them
get exposed to it.
Nonetheless,
I was terribly struck by how far we have come in this
field. I know some people say we have really not moved,
and they have a point. For example, the force field
methods we use today have not changed much since the
early 1990s; there has not been a conceptual
breakthrough to handle solvation in a better way or to
deal with polarization effects upon binding of a small
molecule to its receptor.
People keep to given simplified formalisms, and
there has been little movement towards a better
understanding of many important physicochemical effects.
We
really need to hear more about physical properties of
compounds and how they interact with targets. The
analysis by one speaker of how the dipolar field in a
cavity could affect the binding of a small molecule was
interesting; such analyses need to be much better
elaborated in many cases.
Still,
the drug discovery field has moved tremendously, but we
desperately need much more experimental data out there.
Although there is already an enormous amount of
data, many algorithms, and a lot of understanding, there
are a lot of white areas that have to be filled in the
molecular property and interaction space.
There is no public funding for this type of
research, so companies have to do it in-house. Indeed, a
lot of data, such as pharmacokinetic or physicochemical
properties, are obtained only inside companies and the
academics have no access to it.
Murcko:
Every one of us has built a properties group that is
gathering that kind of information around particular
projects.
Müller:
So everyone has had to do a lot of work to fill the
white space around molecules of prime interest. And
that’s all proprietary, and academia should be jumping
in but they really can’t because that information will
often be published many years later, if at all. And to
people at the cutting edge of science, it may look
boring to gather all these physical properties.
Murcko:
We generate complex thermodynamic and other
“boring” information, under a wide range of
conditions — we do all that because it helps us
discover new medicines.
But in academia it is not funded and it is not
“exciting” enough.
Müller:
Indeed, it has been a very slow moving process.
Meanwhile the structural genomics initiatives
tend to focus on the low hanging fruits in structural
biology. The
really interesting cases, however, the ca. 30% membrane
proteins, which are very difficult to handle, but are
most interesting for the pharmaceutical industry, have
not been tackled by the initiative, but left to industry
to struggle with.
Murcko:
We have to make a distinction.
A lot of people think structure-based design is
for designing a nanomolar inhibitor using the crystal
structure of the target of interest.
But we are not just attempting to design
inhibitors — we want drug candidates.
We need all the other kinds of data — physical
properties, PK, cell data, pharmacology, and so forth
— to help us do that.
So structure-based design means the whole process
of drug discovery being illuminated by what the crystal
structures can teach us about the kinds of molecules
that will “fit” in the target of interest.
Müller:
The sobering aspect of this meeting, and of so
many others of this kind, is that it is so focused on
just this part — the structure-based ligand design.
Many people are happy when they have a nanomolar
binding compound. Typically “structure-based” means
you know the three-dimensional structure of your target,
and use it in the design of a small binding molecule. I
emphasize the term “property-based” to describe the
process we are engaged in, where you are also focusing
on the properties of your molecule and try to use this
as the rational basis in the design of your ligand, so
it will not only fit to the target, but also have the
right spectrum of physicochemical and pharmacological
properties.
Murcko:
For example, say a molecule binds to one of the
P450 isozymes, or to serum albumin.
What do those binding phenomena look like?
How can we use this information to eliminate
potential problems?
Müller:
In addition, academics often don’t care much
about the relevant value range of a given parameter.
Take for example protein binding.
Typically, academics would examine a range from
60% up to 100%, but this is not relevant to us.
What we are interested in is the narrow window
between, say, 98% and 100%. This is much more
challenging for any theoretical prediction method, but
it is crucial for us.
Again, most of that data is in-house and
proprietary. Out of this discussion could perhaps come a
way to set up a publicly available database with data
relevant in drug discovery that could be used by anyone
in academic groups to gear their models to relevant
data.
Tomi
Sawyer: But then
you need to have common protocols or these physico-chemical
data are inconsistent. We’re using different standards
for things like solubility, and so there is no
comparability.
Müller:
Yes, it has been a big effort for us just to make such
data consistent across all Roche Research Centers
worldwide, let alone among us and outside groups.
Sean
Ekins: One of the
issues that came up during the meeting roundtable, was
how we can share structure data and more broadly,
biology data. I
think the NIH Chemical Biology initiative is one start.
It is something that is already there.
That particular initiative looks compelling,
although they need to get enough data in there to get a
critical mass. But
that type of initiative does need to start somewhere.
Murcko:
Another problem is that with many of the papers
that are published, the authors are either not using
good internal controls or they are not properly
validating the data.
We spend a lot of time trying to replicate
published computational methods, and I can tell you that
a very large proportion of them just do not work as
advertised, or only work on the small test case in the
paper.
Müller:
But a lot of people are doing the best they can,
and it’s partly our job to sort it out and publish
what really works. Things
are slowly improving, I think some of the start-ups
still live in an ivory tower — if it works within
their confines, they consider that it works in general.
We have to do our own feasibility studies and
analyses when we are considering any new technology.
The
critical thing is that information we have about protein
structures has improved by leaps and bounds.
The information about soluble proteins has
increased exponentially, while that about membrane-bound
proteins is still far behind and not nearly sufficient.
We also need a lot more information about how
small molecules interact with proteins.
With
structure prediction, you find that there is progress
but again, it is limited. We can predict structure, but
can we predict folds based on sequence?
Can we predict the loops in GPCRs?
At least one company says they can, but it’s
yet to be proved. And, it’s possible that the loops
are just too flexible, maybe they can’t really be
predicted.
Murcko:
Then it gets tougher as you try to predict
physical properties – solubility, HERG binding, etc.
If you ask will this molecular have a solubility
above or below a certain level, you would be right by
random chance about ½ the time. But we need to be right
at least 80% of the time if we are going to really
improve things on the chemistry side.
If you can eliminate a significant fraction of
the molecules the chemists have to make in any series
right off the bat, that would really help.
Sawyer:
What is needed is some sharing of data. What
would be really powerful is a consortium that involved
industry “liberating” some information.
I think some industry folks need to convince
management to give something up as it is a noble cause
for science.
Murcko:
There is a good analogy in SEMATECH, which was
formed in about 1987. That was a consortium of
semiconductor companies who decided to share
pre-competitive data.
They all paid an entry fee and they all joined
because they felt threatened — these US companies were
losing market share to the Japanese — and out of
desperation a group of major players finally took the
plunge. Once
they started, it took at least two years before they
could even learn how to share.
It was only successful because some of the most
influential people in the industry, including Gordon
Moore, were involved.
And, because it was a time of desperation,
they got strong support from the most senior people in
every company. It
is interesting to contemplate whether pharmaceutical
companies could do the same thing around pre-competitive
data to determine what are the best tools to design
drugs. I
believe we must pool information.
Ekins:
Everyone is very focused, and worried, about the
data, when what we should be sharing is the tools. What
are the best high throughput tools? What are the best
approaches to clinical trials?
Murcko:
That is actually what SEMATECH did. It was not
focused as much on the analysis of data, as it was on
using the data to understand which are the best
processes.
Sawyer:
For this to happen, pharma would have to take the
role of captain, and companies would have to feel that
not only was there something in it for them, but that
there was essentially no risk associated with it.
The most valuable data will never be shared, but
if you can get some honest data that would help.
Murcko:
That may be a long way off, but what I find encouraging
is that as we get more structures, especially
protein-protein complexes — we are starting to
understand the rules better, and the subtleties of
biological interactions.
Some of this is already opening up new avenues
for drug discovery.
|
